Manipulation Detection in Satellite Images Using Deep Belief Networks

Janos Horvath, Daniel Mas Montserrat, Hanxiang Hao, Edward J. Delp; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 664-665

Satellite images are more accessible with the increase of commercial satellites being orbited. These images are used in a wide range of applications including agricultural management, meteorological prediction, damage assessment from natural disasters and cartography. Image manipulation tools including both manual editing tools and automated techniques can be easily used to tamper and modify satellite imagery. One type of manipulation that we examine in this paper is the splice attack where a region from one image (or the same image) is inserted ("spliced") into an image. In this paper, we present a one-class detection method based on deep belief networks (DBN) for splicing detection and localization without using any prior knowledge of the manipulations. We evaluate the performance of our approach and show that it provides good detection and localization accuracies in small forgeries compared to other approaches.